Microsoft Nscale UK AI Data Center Deal Withdrawal

Microsoft was set to anchor Nscale's planned AI supercomputer in Loughton, Essex, renting 23,040 Nvidia GB300 GPUs starting in the first quarter of 2027. It walked away after UK Power Networks told Nscale the site's grid connection might not arrive until the early-to-mid 2030s - nearly a decade later than planned ^[1]. Capacity Media's analysis frames this as proof that the limiting factor on hyperscale AI buildouts has shifted from chip supply to electricity: a 23,000-GPU deployment due in 2027 simply cannot wait for a grid connection pencilled in for the next decade ^[2]. On Reddit, the top-voted read of the story was similarly mundane - Microsoft wanted the capacity, could not get it delivered on an acceptable timeline, and left, with commenters pointing to other unrelated UK and European projects facing multi-year grid waits of their own. A vocal counter-view pushed back on any reading that frames this as Microsoft quietly cutting AI spending, arguing the walk-away only makes sense if the power genuinely was not coming.

The withdrawal lands weeks before Nscale's planned US listing, which could raise up to $3 billion at a reported valuation of up to $35 billion ^[2]. That timing matters because Microsoft and Anthropic together account for roughly 85% of Nscale's contracted revenue, with Microsoft alone representing more than 40% of a reported $103 billion backlog ^[3]. Anthropic's own Nscale agreement - a separate $45 billion, six-year deal for about 460MW of capacity at a West Virginia campus - is untouched by the UK pullout ^[4], but the Loughton exit still hands prospective IPO investors a concrete example of how quickly a single-site problem can erase a chunk of forward revenue, and Capacity Media's analysis explicitly questions how firm the rest of that two-customer backlog really is ^[2].

Commentators drew a direct line from Loughton to CoreWeave and the wider neocloud sector. Industry commentator Bertrand Duperrin argues that the risk for compute providers like CoreWeave is not only financial but operational: 'The risk for CoreWeave is therefore not only financial but also operational. Contracts must translate into available capacity, which in turn must be powered.' ^[5]Loughton is a concrete illustration of that argument - a fully contracted, fully financed data center that still cannot open because the power was never actually secured, regardless of how firm the paperwork looked.

Both companies were careful to frame this as narrow. Microsoft said: 'We remain committed to our partnership with Nscale, including work to build robust digital infrastructure and support AI growth globally. This decision relates to a single site.' ^[6]Nscale echoed that framing, calling it 'this unfortunate London-wide grid issue' with 'no impact on our other contracts,' and said it remains committed to Loughton while exploring alternative power sources and other UK expansion sites ^[1]. The story was first reported by The Telegraph and confirmed by Bloomberg ^[7]days ahead of the IPO. Even so, reaction across social platforms skewed skeptical rather than reassured - sentiment converged on the same read as Capacity Media: that the real story is less about one grid connection and more about whether a neocloud's contracted backlog means anything until it is actually powered.